From its earliest days, artificial intelligence (AI) has captivated and enticed the business world with its potential ability to learn not only to imitate humans but to supersede our capabilities. As the importance of digital transformation grows, so too has the number of organizations implementing AI technologies to optimize and automate their business processes.
Process automation and data analytics powered by machine learning are well-established uses for artificial intelligence in today’s marketplace. While these technologies certainly create value and cut costs for companies large and small, we have not yet reached the pinnacle of AI’s potential benefits. Deep reinforcement learning (DRL) — an established technology only now being put to widespread use — may take business process optimization to new heights by leveraging iterative self-improvement to achieve strategic, creative solutions.
What Is Deep Reinforcement Learning?
AI already powers a number of common technologies in use today. Robotic process automation streamlines high-volume tasks and eliminates human errors. Machine learning algorithms mine historical data to create predictive models, identify patterns and glean strategic insights.
While it does use iterative testing to improve performance, DRL is different from its less sophisticated siblings. Rather than generating potential outcomes based on historical data, deep reinforcement learning teaches AI agents and machines with the time-tested “carrot and stick” method. It is, in effect, a direct application of the Markov-Decision process.
In a reinforcement learning model, AI agents are placed into simulated environments, where they are presented with two choices guided by established policy. The AI chooses an action and receives either a “penalty” or a “reward” (e.g., gaining or losing points on an internal scorecard). Positive behaviors are reinforced; negative behaviors are discouraged. In this phase, the AI is performing exploration via trial and error.
Whether its choice yields positive or negative results, the AI updates its policy based on the outcome and then begins the process again with updated choices, this time guided by the updated policy. These subsequent iterations are referred to as “exploitation” as the AI now puts what it has learned to use in seeking an optimal outcome.
The huge advantage here is that AI can iterate processes many times faster than humans can, yielding more useful results more quickly. DRL can empower AI systems to break down and optimize very dynamic, contingency-based processes — particularly if multiple agents work in concert. For example, companies could optimize pricing based on real-time supply and demand data, or optimize warehouse storage and logistics by analyzing distribution patterns, available space as well as boxing, routing and delivery paths.
Deep Reinforcement Learning In Action
The hypothetical applications of DRL are potentially limitless. But companies are already testing the practical applications in a number of ways.
• AI is using deep reinforcement learning to teach autonomous vehicles to drive in about a day.
• Google uses technology developed by DeepMind to actively monitor and cool its data centers for optimal performance and energy efficiency.
• Scientists at Ohio State University, Microsoft and Stanford University are collaborating to improve dialogue generation technology that could lead to smarter, more helpful chatbots and automated assistants.
• Spotify’s AI uses a combination of deep reinforcement learning techniques to curate playlists for millions of users and recommend songs listeners will want to hear, improving customer experience and building brand loyalty.
• The Royal Bank of Canada’s I. trading platform, Aiden, uses deep reinforcement learning to conduct stock market trades based on a range of dynamic inputs, including market conditions, newly discovered trade relationships, etc.
How To Put DRL To Use In Your Business
Is DRL the next step in your company’s digital transformation journey? Answering a few questions can help you decide:
1. What problems are we trying to solve?
Knowing where you are is essential to getting where you want to be. Take an inventory of multi-step business processes and the desired outcomes for each. Deep reinforcement learning works best for processes with numerous, high-volume actions and readily observable outcomes. It’s also very useful if you do not have an existing trove of historical information to train an AI. Not all problems are a nail waiting for the hammer of DRL; simple processes such as invoice processing can be more easily handled with basic robotic process automation, for example.
2. Are we ready to leverage DRL effectively?
In order to achieve an optimal ROI, it is important to have stakeholders for each process collaborate closely with your IT and tech partners to develop policy, develop environments and designate the actions and rewards for each associated agent.
It pays to start small. Limit the scope to the smallest reasonable data sets to achieve a useful outcome and expand as needed. Complexity can grow along with the sophistication of the agents over time. Be ready to balance short- and long-term benefits and carefully consider codependencies and contingencies that might impact the process and, by extension, the AI’s ability to provide useful output.
3. Do we have the tech stack to handle DRL?
Even if you’ve already started your digital transformation initiative, you may need to make significant investments to support DRL and put it to proper use. Can your systems support heavy A/B testing for customer-facing chatbots, for example? Do you have the resources to create multiple test environments to fine-tune AI agents? What are your target timeframe and ROI for implementation of DRL technology in your systems?
Patience and a strategic implementation that fits your budget and your goals will ensure you get the biggest bang for your deep reinforcement learning buck.
As technology continues to evolve, humans will naturally look for ways to leverage those improvements to their benefit. Through iterative optimization, DRL has the potential to help businesses increase the value they’ve gained from digital transformation technologies and take advantage of new opportunities to improve their strategic and competitive advantage.
This article has been published from the source link without modifications to the text. Only the headline has been changed.
